Haiping Huang - Statistical Mechanics of Neural Networks

#book

NotebookLM from Google has new feature to make a podcast from provided resources, it's pretty cool to try.

Video for more information.

Daily dose of data science archive 2024

How to use open wbui and run LLM models locally:
The cleanest method is to use docker, So if you have installed docker on your local machine and also have installed olama just need to download the image from docker with the following command:

docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main

find more options from here.

I uploaded a 360 page book and asked some questions, seems work fine.
Every thing runs locally without need to connect internet after getting the models.

Demo credit: Tarfandoon

Equivalently or even easier :

https://lmstudio.ai/

I am still trying this one, Got some issue reading articles in PDF. Only read Citation part :| .

Copilot now has a free version. unfortunately need vpn from Iran, we are banned 🤷‍♂.

Link

https://centuri-livingsystems.org/recruitment/

PhD and postdoc Applications Now Open
The deadline for submission is January 27.
CENTURI, Marseille, France

#j2p: A simple Python package to convert Jupyter notebooks to Python scripts.

If you find yourself needing to convert a notebook to a Python script, you likely turn to nbconvert. However, this often results in a script with annoying cell separators. Consequently, you may try manually removing these extra lines to focus solely on the code itself.
This tiny package provide a cleaner solution

## Installation

pip install ju2py

## Usage

j2p example.ipynb [output.py]

output name is optional.

P.S:
There is already a package (not by me) for the reverse action

pip install p2j
p2j example.py

GitHub: https://github.com/Ziaeemehr/j2p

Our new research article from PNAS investigates how localized brain manipulations, such as lesions or silencing, impact the entire brain's functional connectivity in mice. Combining fMRI data with computational modeling, the study reveals that these targeted interventions lead to widespread network reconfigurations, sometimes decreasing and other times increasing connectivity. We used personalized brain simulations to explore the underlying mechanisms of this phenomenon, known as diaschisis, finding that alterations in local neuronal excitability drive these global changes. The findings offer insights into understanding the broad effects of focal brain disruptions and could inform the development of more precise therapeutic strategies targeting brain dynamics. The data and analysis tools are publicly available.

https://www.pnas.org/doi/10.1073/pnas.2405706122

ام‌اس (Multiple Sclerosis) یک بیماری خودایمنی است که سیستم عصبی مرکزی را درگیر می‌کند و منجر به ضایعاتی در غلاف میلین می‌شود. این آسیب به میلین باعث کند شدن سرعت هدایت سیگنال‌های عصبی می‌شود که به آن تاخیر هدایتی می‌گویند.
هدف اصلی این کار، برآورد ارتباط بین شدت ضایعات میلین در هر بیمار و افزایش ناشی از آن در تاخیرهای هدایتی در سراسر راه‌های عصبی آسیب‌دیده بود. چگونگی ترجمه دقیق شدت ضایعات ساختاری میلین به کند شدن تاخیرهای هدایت عصبی تاکنون ناشناخته بود.
در این مطالعه از داده‌های ۳۸ نفر (۲۰ فرد سالم و ۱۸ بیمار مبتلا به ام‌اس) استفاده کردیم که شامل ثبت فعالیت مغزی با مگنتوانسفالوگرافی (MEG) و تصویربرداری رزونانس مغناطیسی (MRI) برای تحلیل ساختار مغز و ضایعات ماده سفید بود.
همچنین از مدل‌های محاسباتی بزرگ‌مقیاس مغز و تکنیکی به نام استنتاج مبتنی بر شبیه‌سازی (Simulation-Based Inference - SBI) استفاده شده است.
ادامه …
LinkedIn

2025-06-12 : Workshop on Model Inversion
When: Thursday June 12th 14:00 to 18:00Where: Salle Laurent Vinay, Institut de Neurosciences de la Timone, Marseille, France.

Page Link: https://conect-int.github.io/
Zoomlink: https://univ-amu-fr.zoom.us/j/98265637982?pwd=H3XzYziirf301CBX327rFFaDbCKHW4.1

Dear all,
Have you ever asked yourself how to find the neural model that best describes your data? What a good question! For complex models, no easy solution exists. Generally, this issue is referred to as "model inversion", and it often represents an ill-posed problem in data science, where no unique solution is at hand. However, recent advances in ML and AI are providing interesting tools that can be used to perform model inversion and fit neural models to brain data.
The aim of the workshop is to provide an overview of projects focusing on model inversion. Although technical, the workshop will try to provide an overview for experimentalists and those who are not familiar with model inversion techniques.

PROGRAM

12 June 2025 (Salle Laurent Vinay, INT)
14:00 Nina Baldy (TNG-INS) - Dynamic Causal Modeling in Probabilistic Programming Languages14:45 Pedro Garcia (BraiNets-INT) - A Dynamic Causal Model to infer effective connectivity from meg induced responses (high-gamma-activity): a workflow for model bayesian inversion
15:30 Pause coffee: :mate_drink:
15:45 Cyprien Dautrevaux (BraiNets-INT) - Dynamic Causal Modelling for ERPs propagation estimated from MEG
16:30 Jean-Didier Lemaréchal (BraiNets-INT) - Bayesian inference applied to neuronal models: methods & applications
17:15 Abolfazl Ziaeemehr (TNG-INS) - Virtual Brain Inference (VBI): A flexible and integrative toolkit for efficient probabilistic inference on virtual brain models

PhD #Position
https://elifkoksal.github.io/positions.html

Multiscale brain rhythms under healthy and epileptic conditions: computational modeling insights for clinical applications

Neural activity in the brain operates across multiple scales, encompassing both spatial and temporal dynamics. In patients with epilepsy, however, cognitive impairments are often linked to disruptions in these neural mechanisms, particularly through interictal epileptiform discharges (IEDs). This project aims to uncover new insights into the link between electrophysiology and attention deficits, one of the most prevalent cognitive impairments in patients with epilepsy, by exploring the role of IEDs. The PhD candidate will develop a comprehensive neocortical population model. The model will be validated on electrophysiological signals recorded in epileptic patients, and its dynamics will be studied to detail the mechanisms of multiple timescale interactions giving rise to healthy and pathological activity.

The research project is at the interface between computational, cognitive, and clinical neurosciences. The candidate will preferably have some background in applied mathematics or computational neuroscience/systems biology. Programming skills in Python and knowledge of dynamical systems are required. Knowledge in cognitive neuroscience, electrophysiology and/or EEG analysis would be an asset. The PhD fellow will join the Cophy Team hosted at the Center for Neuroscience Research of Lyon (CRNL), France. The ideal start date is September 2025, with some flexibility.

Candidates should send their CV, a motivation letter, contact information for 2-3 references and their master degree notes (if available) to Elif Köksal-Ersöz elif.koksal@inria.fr and Mathilde Bonnefond mathilde.bonnefond@inserm.fr until June 10th 2025.